An Efficient Photovoltaic Power Forecasting using Adaptive Learning Rate Enhanced Gated Recurrent Unit (ALRE-GRU) network optimized with Enhanced Dynamic Grasshopper Optimization Algorithm (EDGOA)

Abstract Accurate photovoltaic (PV) power forecasting is essential for the reliable operation of renewable energy systems due to the nonlinear and non-stationary nature of solar power generation. This study proposes a hybrid framework that integrates Variational Mode Decomposition (VMD) with an Adaptive Learning Rate Enhanced Gated Recurrent Unit (ALRE-GRU) network optimized using the Enhanced Dynamic Grasshopper Optimization Algorithm (EDGOA). VMD is applied to decompose PV power signals into multiple band-limited intrinsic modes, improving data quality through noise reduction and feature separation. The decomposed modes are used as inputs to the ALRE-GRU model, which employs a parameter-wise adaptive learning rate strategy based on first- and second-order gradient moments to enhance training stability and convergence speed. EDGOA is further utilized to optimally tune critical hyperparameters, particularly the hidden layer size. The incorporation of dynamic inertial motion weights and an adaptive mutation probability coefficient within the Enhanced Dynamic Grasshopper Optimization Algorithm (EDGOA) effectively mitigates the risk of premature convergence to suboptimal local optima. The proposed method is evaluated using real-world PV data from the Yulara Solar System in Australia for short-term (60-hour) and long-term (300-hour) forecasting. Experimental results demonstrate superior performance, achieving an R² of 99.91%, RMSE of 11.41, MAE of 8.51, and MAPE of 0.70%, outperforming several state-of-the-art forecasting models. These results confirm the effectiveness and robustness of the proposed framework for photovoltaic power forecasting applications.